import tensorflow as tf
from einops import rearrange
from fusionlab.functional.tfdice import tf_dice_score
__all__ = ["TFDiceLoss", "TFDiceCE"]
BINARY_MODE = "binary"
MULTICLASS_MODE = "multiclass"
# TODO: Test code
[docs]
class TFDiceCE(tf.keras.losses.Loss):
[docs]
def __init__(self, mode="binary", from_logits=False, w_dice=0.5, w_ce=0.5):
"""
Dice Loss + Cross Entropy Loss
Args:
w_dice: weight of Dice Loss
w_ce: weight of CrossEntropy loss
mode: Metric mode {'binary', 'multiclass'}
"""
super().__init__()
self.w_dice = w_dice
self.w_ce = w_ce
self.dice = TFDiceLoss(mode, from_logits)
if mode == BINARY_MODE:
self.ce = tf.keras.losses.BinaryCrossentropy(from_logits)
elif mode == MULTICLASS_MODE:
self.ce = tf.keras.losses.SparseCategoricalCrossentropy(from_logits)
[docs]
def call(self, y_true, y_pred):
loss_dice = self.dice(y_true, y_pred)
loss_ce = self.ce(y_true, y_pred)
return self.w_dice * loss_dice + self.w_ce * loss_ce
[docs]
class TFDiceLoss(tf.keras.losses.Loss):
[docs]
def __init__(
self,
mode="multiclass", # binary, multiclass
log_loss=False,
from_logits=False,
):
"""
Implementation of Dice loss for image segmentation task.
It supports "binary", "multiclass"
https://github.com/BloodAxe/pytorch-toolbelt/blob/develop/pytorch_toolbelt/losses/dice.py
Args:
mode: Metric mode {'binary', 'multiclass'}
log_loss: If True, loss computed as `-log(dice)`; otherwise `1 - dice`
from_logits: If True assumes input is raw logits
"""
super().__init__()
self.mode = mode
self.from_logits = from_logits
self.log_loss = log_loss
[docs]
def call(self, y_true, y_pred):
"""
:param y_true: (N, *)
:param y_pred: (N, *, C)
:return: scalar
"""
y_true_shape = y_true.shape.as_list()
y_pred_shape = y_pred.shape.as_list()
assert y_true_shape[0] == y_pred_shape[0]
num_classes = y_pred_shape[-1]
axis = [0]
if self.from_logits:
# get [0..1] class probabilities
if self.mode == MULTICLASS_MODE:
y_pred = tf.nn.softmax(y_pred, axis=-1)
else:
y_pred = tf.nn.sigmoid(y_pred)
if self.mode == BINARY_MODE:
y_true = rearrange(y_true, "... -> (...) 1")
y_pred = rearrange(y_pred, "... -> (...) 1")
elif self.mode == MULTICLASS_MODE:
y_true = tf.cast(y_true, tf.int32)
y_true = tf.one_hot(y_true, num_classes)
y_true = rearrange(y_true, "... C -> (...) C")
y_pred = rearrange(y_pred, "... C -> (...) C")
else:
AssertionError("Not implemented")
scores = tf_dice_score(y_pred, tf.cast(y_true, y_pred.dtype), axis=axis)
if self.log_loss:
loss = -tf.math.log(tf.clip_by_value(scores, clip_value_min=1e-7, clip_value_max=scores.dtype.max))
else:
loss = 1.0 - scores
return tf.math.reduce_mean(loss)
if __name__ == '__main__':
print("Multiclass")
pred = tf.convert_to_tensor([[
[1., 2., 3., 4.],
[2., 6., 4., 4.],
[9., 6., 3., 4.]
]])
pred = rearrange(pred, "N C H -> N H C")
true = tf.convert_to_tensor([[2, 1, 0, 2]])
dice = TFDiceLoss("multiclass", from_logits=True)
loss = dice(true, pred)
print("Binary")
pred = tf.convert_to_tensor([0.4, 0.2, 0.3, 0.5])
pred = tf.reshape(pred, [1, 2, 2, 1])
true = tf.convert_to_tensor([0, 1, 0, 1])
true = tf.reshape(true, [1, 2, 2])
dice = TFDiceLoss("binary", from_logits=True)
loss = dice(true, pred)
print("Binary Log loss")
pred = tf.convert_to_tensor([0.4, 0.2, 0.3, 0.5])
pred = tf.reshape(pred, [1, 2, 2, 1])
true = tf.convert_to_tensor([0, 1, 0, 1])
true = tf.reshape(true, [1, 2, 2])
dice = TFDiceLoss("binary", from_logits=True, log_loss=True)
loss = dice(true, pred)